the labour supply effect of the abolition of the earnings rule for older workers in the united kingdom

33 Pages·2002·0.68 MB·English

Working Paper 17/02

THE LABOUR SUPPLY EFFECT OF THE ABOLITION OF THE
EARNINGS RULE FOR OLDER WORKERS IN THE UNITED
KINGDOM

*
Richard Disney
†
Sarah Smith

* University of Nottingham, and Institute for Fiscal Studies, London. Corresponding author: School of Economics,
University of Nottingham, NG7 2RD, UK; email: [email protected]

† Financial Services Authority
October 2001
THE LABOUR SUPPLY EFFECT OF THE ABOLITION OF THE
EARNINGS RULE FOR OLDER WORKERS IN THE UNITED
KINGDOM

by
Richard Disney∗ and Sarah Smith†

Abstract
High effective tax rates on work at and around state pension age deter
participation. An example is the ‘earnings test’ operating in several OECD
countries. The United States abolished its test for the 65+ age group in 2000.
The United Kingdom offers a ‘natural experiment’ of this reform, as it abolished
its test, known as the ‘earnings rule', in 1989. We examine the effect of this
change, taking account of the opportunity to defer pension rights. Abolition of
the rule raised working hours of older male workers by around 4 hours a week,
with a lesser impact on women’s behaviour.

JEL Classification: J22 J26
Key Words: Labour supply; Retirement; Earnings Test
Acknowledgements
Funding for this research has been provided by the ESRC Centre for the Microeconomic
Analysis of Fiscal Policy. Material from the Family Expenditure Survey was made available
by the ONS through the ESRC Data Archive and has been used by permission of the
controller of ONS. The authors wish to thank two referees, Richard Blundell, Barry
Bosworth, Gary Burtless, Bill Gale, Amanda Gosling, Jon Gruber, Richard Johnson, Costas
Meghir, John van Reenen, Edward Whitehouse and other seminar and workshop participants
at IFS, the IMF, the University of Manchester, a TMR Savings and Pensions conference and
the World Bank for comments. The normal disclaimer applies. In addition, the Financial
Services Authority is not responsible for any views expressed here.
THE LABOUR SUPPLY EFFECT OF THE ABOLITION OF THE
EARNINGS RULE FOR OLDER WORKERS IN THE UNITED
KINGDOM

1. Introduction

The decline in hours of paid work among the elderly, primarily among older
men, in OECD countries is well documented (for example, OECD, 1996) The United
Kingdom is no exception to this trend – see, for example, Blundell and Johnson
(1999) and Disney (1999). An influential school of thought links this decline, and the
relative speed of decline in different countries, to the magnitude of effective taxes on
remaining in work for older people (Gruber and Wise, 1999). Reduce these effective
taxes, it is argued, and workers will defer retirement until a later date.
A particularly extreme version of a tax on older workers are the earnings tests
which operate in a number of OECD countries.1 These restrict the amount of state
pension that can be received by people who continue to work past pensionable age, by
withdrawing the pension in line with earnings at high marginal rates. Table 1 depicts
the nature of these tests for OECD countries. As can be seen, there is a wide variety
of practice, from environments where it is basically impossible simultaneously to
work and to receive a public pension (such as Ireland, Portugal and Spain) through to
regimes where earnings can be received without extra penalty (of course, such
countries will still typically levy income tax and, in some cases, payroll taxes on post-
retirement earnings). Note also that many countries which operate such tests also
disregard a certain level of earnings in applying the test, and that some countries also
permit individuals to defer their pension, with a higher rate of pension being paid
1
when the individual finally stops working, reducing the test's effective penalty on
working beyond pension age.
< Table 1 here >
The earnings test has received particular prominence recently, with the United
States abandoning the application of its test to people aged over 65 from 2000, with
the general aim of increasing hours of work of older people.2 The debate preceding
the reform generated a certain amount of empirical work designed to simulate the
impact of the policy. The United Kingdom, however, offers a ‘natural experiment’ by
which the effect of such a change can be examined, since it abolished its own earnings
test, known as the ‘earnings rule’ in 1989 (Whitehouse, 1990). A sufficient time
interval has elapsed to permit a proper evaluation of the impact of this change. The
purpose of the present paper, therefore, is to estimate the impact of the abolition of the
earnings rule in the United Kingdom on the hours of work of older workers,
comparing our findings with simulations of similar changes both in the UK and
elsewhere and with one actual study of an abolition, for Canada.
The structure of the remainder of the paper is as follows. Section 2
summarises the operation of the earnings rule in the UK up to 1989 and describes
other studies of earnings tests both in the UK and elsewhere. Section 3 writes down a
simple formal model of the joint decisions concerning retirement and deferral facing
an individual. Section 4 provides empirical evidence on the impact of abolition and
provides a brief conclusion.

2
2. Earnings Tests: basics and empirical methods

2.1 The basics of the earnings test

The imposition of an earnings test generates a highly non-linear static budget
constraint facing workers on reaching pensionable age. Fig. 1 illustrates a budget
constraint that exhibits the main features of the ‘earnings rule’ as it operated in the
United Kingdom until 1989. It assumes a wage rate of £3.50 an hour, a basic state
pension of £44 per week (the level in 1989), a tax rate, net of the age allowance
(assumed equal to the basic pension), of 25% and an earnings rule operating in the
following manner: gross earnings of £75 per week are exempt, the basic pension is
withdrawn at 50 pence per £ of earnings from £75 up to £79, and £ for £ thereafter.3
< Fig. 1 here >
As Fig. 1 demonstrates, the static budget constraint exhibits a convex kink at
the exempt amount (the earnings disregard) at point B and a non-convex kink where
the state pension entitlement is exhausted, at point C. Friedberg (1998) provides
illustrations of the very similar US earnings test, and there are also parallels with the
operation of other benefit regimes, notably those for in-work benefits to low income
families such as the Working Family Tax Credit in the UK (Blundell et al, 2000) and
the Earned Income Tax Credit in the US (Eissa and Liebman, 1996).
In the simplest interpretation, abolition of the earnings rule linearises the
budget constraint, as indicated by the dotted line in Fig. 1. Basic labour supply theory
would then suggest no change in hours for non-participants at point A or for
participants between points A and B, a potentially large increase in hours supplied by
individuals bunched at the kink point B, and a possible negative impact on hours of
3
those had previously worked at some point above C.4 On balance, however, the net
effect of these changes, although depending on the overall distribution of hours, might
be positive, especially where hours are organised in discrete ‘packages’ of part-time
and full-time work (Hurd, 1996). Specifically, abolition of the earnings rule might
encourage individuals who had gone part-time at state pension age in order to avoid
the impact of the rule to choose instead to remain in full time work for a longer
period.
There are, however, complications arising from the earnings rule. First,
individuals in the United Kingdom had, and still have, the opportunity to defer
pension receipt. The earnings rule operated for five years after state pension age (65
for men, 60 for women) until state ‘retirement age’ (70 for men, 65 for women).
Anyone could choose to defer receipt of the state pension for up to five years and
thereby accrue additional pension entitlements at a rate of 7.5% (in 1989) for each
year that they deferred.5 This rate is approximately actuarially unfair for a single man
and actuarially favourable for a single woman, given life expectancies of respectively
14.3 years at 65 for a man and 22.1 years at 60 for a women, although one should also
take account of time discounting, the inheritance of deceased spouse’s benefits, and
the likelihood of self-selection of deferrers on the basis of individual differences in
expected longevity. Note that deferral requires an explicit decision in the UK, whereas
in the US, benefits in each year after normal retirement age are automatically
increased by 6.67% for each year of full benefits lost due to the earnings test through
the Delayed Retirement Credit (Gruber and Orszag, 2000). In Fig. 1, actuarially
unfair deferral is illustrated by the bold dashed line. For such people who chose to
defer, abolition of the earnings rule should act as a pure positive income effect, which
should, on balance, reduce their hours of work.
4
The other complication concerns couples versus single people. Reforms of
this kind may also affect spouses’ behaviour, especially where one partner is over
state pensionable age and the other is below. For example, abolition of the earnings
rule might induce the older partner to continue in full time work rather than switch to
part time work, which might in turn affect the hours or participation decision of the
younger partner.6
2.2 Previous research on earnings tests

In contrast with other areas of labour supply, the literature on earnings tests
and older workers is limited. Moreover, much of the literature does not adequately
take adequate account of the deferral option or of the problem of handling the
behaviour of spouses. The traditional approach in the United States, exemplified by
Burtless and Moffitt (1985), utilises a piecewise linear budget constraint approach to
estimate labour supply effects of the earnings test in the US for an individual worker.
Typically, such studies conclude that the test has had little effect on labour supply (for
a survey, see Leonesio, 1990).
A criticism of such studies, noted by Friedberg (2000) is that, in the absence
of temporal variations in the tax structure (i.e. changes in the disregards, tax rates or
even outright abolition of the earnings test), estimates depend on cross sectional
variation in components of income such as the wage rate and unearned income. If
these correlate with unobservables, then the estimated effects will be biased.
Gustman and Steinmeier’s study (1986) may be subject to the same criticism
concerning unobservables, but an added reason for their finding of a relatively small
effect lies in their attempt to model the impact of the Delayed Retirement Credit in a
5
structural retirement model which, as we suggested previously, should alleviate the
impact of the test. The idea that an earnings test might affect the timing of the first
claim on social security benefit is explored by Gruber and Orszag (2000) who argue
that abolition of the test might accelerate the first claim on social security, so reducing
labour supply in total irrespective of the behaviour of those round the kink (i.e. point
B in Fig. 1).
A further criticism of quasi-structural estimation of the model (at least of its
static component) arises from the use of the piecewise linear budget constraint method
itself. MaCurdy, Green and Paarsch (1990) argue that the log likelihood is only
defined for individuals locating at a kink like point B in Fig. 1 if the compensated
substitution effect is positive. Thus there is an inherent bias in the method towards
finding positive compensated substitution effects. Blundell et al (1998) circumvent
this issue in another context by dropping observations at the kink and estimating
labour supply elasticities over the rest of the sample, selectivity corrected. As
Friedberg (2000) points out, however, their procedure is unappealing in the context of
the earnings test since much of the ‘action’ is expected to derive precisely from the
behaviour of those at the kink. Her own estimates by the piecewise linear budget
constraint method, she argues, do not require the imposition of a positive
compensated substitution effect and avoid the issue of correlation with unobservables
by exploiting temporal variations in both the level of the disregard (exempt amount)
to the earnings test and changes in the effective tax rate, between 1978 and 1990. She
utilises the resulting elasticities to estimate that abolition of the earnings test in the US
would raise hours worked of those at or above the kink at point B by 5.3%. Put
simply, this would raise the hours of a part timer working 20 hours a week by one
hour.
6
However Friedberg’s work is also not immune from criticism. The first is that
the ‘dynamic’ aspect of the issue – the accrual of Delayed Retirement Credit (DRC)
from any tax levied as a result of the earnings test – is simply ignored on the grounds
that people do not understand the DRC. Effectively, it is assumed that individuals are
sophisticated enough to adjust their labour supply to small variations in exempt
amounts and effective tax rates while systematically disregarding the fact that tax
levied will be recovered later through higher pension benefits. Second, there is always
danger in extending marginal elasticities derived from small changes to construct
hypothetical outcomes for large changes, such as the abolition of a whole tax regime.
The only previous study of the earnings rule in the UK is contained in a
pathbreaking study by Zabalza et al (1980). They utilised data for a cross-section of
people aged between 50 and 73 in 1977 to develop a discrete tri-choice model in
which individuals could choose between full-time and part-time work, and non-
participation. By imposing a CES utility function and requiring convexity of the
opportunity set in the ordering full-time, part-time, non-participation, they were able
to predict chosen states (78% of the time) and to use parameter estimates to simulate
policy changes. Specifically, abolition of the earnings rule would leave participation
unaffected, but raised the average hours worked (averaged over all people) by about
2% for men and 1.6% for women. Since roughly 20% of people in the relevant age
range worked in the late 1970s, this gives somewhat larger magnitudes than
Friedberg’s estimates.
Zabalza et al may however overstate the success of their model. Since the age
range is fairly broad, predictive power is achieved largely by predicting that people
under pension age work and that the majority of older people over pension age do not
work. Only 10% of actual part-time workers (amongst whom are those at the kink at
7
point B in Fig. 1) are successfully predicted by their model. Consequently, the
simulated shift from part-time to full-time work arising from the abolition of the
earnings rule must have a very high standard error.
What characterises all these studies is that they rely on simulated responses
derived from labour supply modelling, although Friedberg (2000), like Blundell et al
(1998), exploits policy variation to identify elements of model structure. A different
empirical strategy is to look at actual reforms – in this case, actual abolition of
earnings tests – to estimate policy effects. A standard approach in this case uses
‘differences of differences’, which requires finding a ‘control group’ who are
unaffected by the reform in question and who are affected identically by other
‘shocks’ (for example, to labour demand). This is the approach used here. It is
surveyed by Angrist and Krueger (1999) and utilised in a number of comparable
policy experiments concerning tax regimes.7
Only one study, to our knowledge, applies the method to an earnings test.
Baker and Benjamin (1999) examine the sequential elimination of earnings tests from
pension plans in Canada in the mid-1970s. They exploit the fact that the Quebec
Pension Plan and the (rest of) Canada Pension Plan abolished their earnings tests at
different times, so giving an overidentifying test. Moreover, there was no equivalent
to the Delayed Retirement Credit in the US to complicate the picture: taxed away
benefits were simply lost.8 This gives a potentially clean test of the impact. Baker
and Benjamin find no evidence that abolition of the tests affected participation, some
evidence that take-up of benefits was affected and, most pertinently, some evidence of
a shift from part-time to full-time work. However this shift in general took the form
of a shift in the number of weeks worked per year (+5 to 6 weeks) rather than a shift
in hours per week. They argue that this result is consistent with a discrete shift fixed-
8